[RFC] Remaining test failures on Windows

We're down to a more reasonable single digit number of failures (though the new CI machine seems to have grown a couple). test-persistent-nodemap-stream-clone.t is likely the most serious, but I've no idea why test-racy-mutations.t loses the last line of hg debugrevlogindex -m on Windows. That's a little concerning. Any ideas about what to check for any of these would be helpful.

This is the collection of tests failing on Windows at this point. There are still some things that I can fix because they are stupid MSYS path mangling issues, but a fair bit of this is dirstate stuff (or other things) that I don't have a lot of knowledge about.

I started the first few commits trying to conditionalize things to run clean on both Windows and Linux, but gave up on that. The next bunch are grouping test changes that look like they're caused by the same thing- the known largefiles dirstate wrapping issue, the debugdirstate output where things are set or unset, and have a mode or not, etc. The 4rd and 5th last are a WIP trying to fill in for the lack of SIGKILL on Windows.

Finally, the last 3 commits are just a collection of the remaining test failures. The 3rd last one was with run-tests.py -r, and then the second last one produced more failures (I guess some things got dropped out of the failed list somehow). The very last commit is a full run over the stack as it is now. These might be best viewed as a diff across all of them, but as noted in the last commit, there are some tests that are non-deterministic for some reason.

@marmoute - mostly this is an FYI for you, and a backup of my WIP. I don't think you'll be able to do much without a Windows machine, but maybe something will jump out, or maybe you'll have guesses I can run through the test suite. Some of the stuff like svn, I don't think is tested on Linux. I'm not even sure exactly what needs to be installed for it, so I'm fine with blacklisting to get off the list for now.